2 * This file is part of the Palacios Virtual Machine Monitor developed
3 * by the V3VEE Project with funding from the United States National
4 * Science Foundation and the Department of Energy.
6 * The V3VEE Project is a joint project between Northwestern University
7 * and the University of New Mexico. You can find out more at
10 * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
11 * Copyright (c) 2008, Lei Xia <lxia@cs.northwestern.edu>
12 * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
13 * All rights reserved.
15 * Author: Jack Lange <jarusl@cs.northwestern.edu>
16 * Lei Xia <lxia@cs.northwestern.edu>
18 * This is free software. You are permitted to use,
19 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
22 #include <palacios/vmm.h>
23 #include <palacios/vmm_dev_mgr.h>
24 #include <palacios/vm_guest_mem.h>
25 #include <devices/lnx_virtio_pci.h>
26 #include <palacios/vmm_vnet.h>
27 #include <palacios/vmm_sprintf.h>
28 #include <devices/pci.h>
31 #ifndef CONFIG_DEBUG_LINUX_VIRTIO_VNET
33 #define PrintDebug(fmt, args...)
37 #define QUEUE_SIZE 4096
38 #define CMD_QUEUE_SIZE 128
44 } __attribute__((packed));
51 struct virtio_vnet_state {
52 struct v3_vm_info * vm;
53 struct vnet_config vnet_cfg;
54 struct virtio_config virtio_cfg;
56 struct vm_device * pci_bus;
57 struct pci_device * pci_dev;
59 struct virtio_queue queue[NUM_QUEUES];
61 struct virtio_queue * cur_queue;
66 ulong_t pkt_sent, pkt_recv, pkt_drop, tx_exit, rx_exit, total_exit;
70 #define VNET_GET_ROUTES 10
71 #define VNET_ADD_ROUTE 11
72 #define VNET_DEL_ROUTE 12
74 #define VNET_GET_LINKS 20
75 #define VNET_ADD_LINK 21
76 #define VNET_DEL_LINK 22
78 // structure of the vnet command header
79 struct vnet_ctrl_hdr {
82 } __attribute__((packed));
85 struct vnet_bridge_pkt {
88 uint8_t pkt[ETHERNET_PACKET_LEN];
89 }__attribute__((packed));
92 static int virtio_reset(struct virtio_vnet_state * vnet_state) {
94 memset(vnet_state->queue, 0, sizeof(struct virtio_queue) * NUM_QUEUES);
96 vnet_state->cur_queue = &(vnet_state->queue[0]);
98 vnet_state->virtio_cfg.status = 0;
99 vnet_state->virtio_cfg.pci_isr = 0;
101 vnet_state->queue[0].queue_size = CMD_QUEUE_SIZE;
102 vnet_state->queue[1].queue_size = QUEUE_SIZE;
103 vnet_state->queue[2].queue_size = QUEUE_SIZE;
105 memset(&(vnet_state->vnet_cfg), 0, sizeof(struct vnet_config));
106 v3_lock_init(&(vnet_state->lock));
113 static int get_desc_count(struct virtio_queue * q, int index) {
114 struct vring_desc * tmp_desc = &(q->desc[index]);
117 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
118 tmp_desc = &(q->desc[tmp_desc->next]);
128 static int handle_cmd_kick(struct guest_info * core,
129 struct virtio_vnet_state * vnet_state) {
130 struct virtio_queue * q = &(vnet_state->queue[0]);
132 PrintDebug("VNET Bridge: Handling command queue\n");
134 while (q->cur_avail_idx != q->avail->index) {
135 struct vring_desc * hdr_desc = NULL;
136 struct vring_desc * buf_desc = NULL;
137 struct vring_desc * status_desc = NULL;
138 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
139 uint16_t desc_cnt = get_desc_count(q, desc_idx);
140 struct vnet_ctrl_hdr * hdr = NULL;
143 uint8_t * status_ptr = NULL;
147 PrintDebug("VNET Bridge: CMD: Descriptor Count=%d, index=%d, desc_idx=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE, desc_idx);
150 PrintError("VNET Bridge cmd must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
154 hdr_desc = &(q->desc[desc_idx]);
156 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
157 PrintError("Could not translate VirtioVNET header address\n");
161 desc_idx = hdr_desc->next;
163 if (hdr->cmd_type == VNET_ADD_ROUTE) {
165 for (i = 0; i < hdr->num_cmds; i++) {
166 uint8_t tmp_status = 0;
167 struct v3_vnet_route * route = NULL;
169 buf_desc = &(q->desc[desc_idx]);
171 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(route)) == -1) {
172 PrintError("Could not translate route address\n");
177 PrintDebug("VNET Bridge: Adding VNET Route\n");
179 tmp_status = v3_vnet_add_route(*route);
181 PrintDebug("VNET Route Added\n");
183 if (tmp_status != 0) {
184 PrintError("Error adding VNET ROUTE\n");
188 xfer_len += buf_desc->length;
189 desc_idx = buf_desc->next;
196 status_desc = &(q->desc[desc_idx]);
198 if (v3_gpa_to_hva(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
199 PrintError("VirtioVNET Error could not translate status address\n");
203 xfer_len += status_desc->length;
204 *status_ptr = status;
206 PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
207 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
208 q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
215 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
216 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
217 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
218 vnet_state->virtio_cfg.pci_isr = 1;
226 static int vnet_pkt_input_cb(struct v3_vm_info * vm,
227 struct v3_vnet_pkt * pkt,
228 void * private_data){
229 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
230 struct virtio_queue * q = &(vnet_state->queue[RECV_QUEUE]);
234 flags = v3_lock_irqsave(vnet_state->lock);
236 if (q->ring_avail_addr == 0) {
237 PrintError("Queue is not set\n");
241 if (q->cur_avail_idx != q->avail->index) {
242 uint16_t pkt_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
243 struct vring_desc * pkt_desc = NULL;
244 struct vnet_bridge_pkt * virtio_pkt = NULL;
246 pkt_desc = &(q->desc[pkt_idx]);
247 PrintDebug("VNET Bridge RX: buffer desc len: %d\n", pkt_desc->length);
249 if (v3_gpa_to_hva(&(vm->cores[0]), pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
250 PrintError("Could not translate buffer address\n");
254 PrintDebug("VNET Bridge: RX: pkt sent to guest pkt size: %d, dst link: %d\n", pkt->size, pkt->dst_id);
256 // Fill in dst packet buffer
257 virtio_pkt->link_id = pkt->dst_id;
258 virtio_pkt->pkt_size = pkt->size;
259 memcpy(virtio_pkt->pkt, pkt->data, pkt->size);
261 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
262 q->used->ring[q->used->index % q->queue_size].length = sizeof(struct vnet_bridge_pkt);
267 vnet_state->pkt_drop ++;
270 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
271 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
272 vnet_state->virtio_cfg.pci_isr = 0x1;
273 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
280 v3_unlock_irqrestore(vnet_state->lock, flags);
285 static int handle_pkt_kick(struct guest_info * core,
286 struct virtio_vnet_state * vnet_state)
288 struct virtio_queue * q = &(vnet_state->queue[XMIT_QUEUE]);
291 if (q->ring_avail_addr == 0) {
295 while (q->cur_avail_idx != q->avail->index) {
296 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
297 struct vring_desc * pkt_desc = NULL;
298 struct vnet_bridge_pkt * virtio_pkt = NULL;
300 pkt_desc = &(q->desc[desc_idx]);
302 PrintDebug("VNET Bridge: Handle TX desc buf_len: %d\n", pkt_desc->length);
304 if (v3_gpa_to_hva(core, pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
305 PrintError("Could not translate buffer address\n");
309 //PrintDebug("VNET Bridge: TX: on cpu %d pkt size: %d, dst link: %d\n", cpu, virtio_pkt->pkt_size, virtio_pkt->link_id);
311 struct v3_vnet_pkt pkt;
312 pkt.size = virtio_pkt->pkt_size;
313 pkt.src_type = LINK_EDGE;
315 memcpy(pkt.header, virtio_pkt->pkt, ETHERNET_HEADER_LEN);
316 pkt.data = virtio_pkt->pkt;
318 v3_vnet_send_pkt(&pkt, NULL);
320 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
321 q->used->ring[q->used->index % q->queue_size].length = pkt_desc->length; // What do we set this to????
324 vnet_state->pkt_sent ++;
334 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
335 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
336 vnet_state->virtio_cfg.pci_isr = 0x1;
342 static void vnet_virtio_poll(struct v3_vm_info * vm, void * private_data){
343 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
345 if(vm == vnet_state->vm){
346 handle_pkt_kick(&(vm->cores[0]), vnet_state);
350 static int handle_rx_kick(struct guest_info *core,
351 struct virtio_vnet_state * vnet_state)
353 //v3_vnet_enable_bridge();
358 static int vnet_virtio_io_write(struct guest_info * core,
359 uint16_t port, void * src,
360 uint_t length, void * private_data) {
361 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
362 int port_idx = port % vnet_state->io_range_size;
364 PrintDebug("VNET Bridge: VIRTIO VNET Write for port %d len=%d, value=%x\n",
365 port, length, *(uint32_t *)src);
367 vnet_state->total_exit ++;
369 case GUEST_FEATURES_PORT:
371 PrintError("Illegal write length for guest features\n");
374 vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
377 case VRING_PG_NUM_PORT:
379 addr_t pfn = *(uint32_t *)src;
380 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
382 vnet_state->cur_queue->pfn = pfn;
384 vnet_state->cur_queue->ring_desc_addr = page_addr ;
385 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
386 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
387 sizeof(struct vring_avail) + \
388 (QUEUE_SIZE * sizeof(uint16_t)));
390 // round up to next page boundary.
391 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
393 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
394 PrintError("Could not translate ring descriptor address\n");
398 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
399 PrintError("Could not translate ring available address\n");
403 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
404 PrintError("Could not translate ring used address\n");
408 PrintDebug("VNET Bridge: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
409 (void *)(vnet_state->cur_queue->ring_desc_addr),
410 (void *)(vnet_state->cur_queue->ring_avail_addr),
411 (void *)(vnet_state->cur_queue->ring_used_addr));
413 PrintDebug("VNET Bridge: RingDesc=%p, Avail=%p, Used=%p\n",
414 vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
416 if(vnet_state->queue[RECV_QUEUE].avail != NULL){
417 vnet_state->ready = 1;
418 vnet_state->queue[RECV_QUEUE].used->flags |= VRING_NO_NOTIFY_FLAG;
421 //No notify when there is pkt tx from guest
422 //palacios will do the polling
423 if(vnet_state->queue[XMIT_QUEUE].used != NULL){
424 vnet_state->queue[XMIT_QUEUE].used->flags |= VRING_NO_NOTIFY_FLAG;
427 PrintError("Illegal write length for page frame number\n");
431 case VRING_Q_SEL_PORT:
432 vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
434 if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
435 PrintError("VNET Bridge device has no qeueues. Selected %d\n",
436 vnet_state->virtio_cfg.vring_queue_selector);
440 vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
443 case VRING_Q_NOTIFY_PORT: {
444 uint16_t queue_idx = *(uint16_t *)src;
446 PrintDebug("VNET Bridge: Handling Kick\n");
448 if (queue_idx == 0) {
449 if (handle_cmd_kick(core, vnet_state) == -1) {
450 PrintError("Could not handle Virtio VNET Control command\n");
453 } else if (queue_idx == 1) {
454 if (handle_pkt_kick(core, vnet_state) == -1){
455 PrintError("Could not handle Virtio VNET TX\n");
458 PrintError("Notify on TX\n");
459 } else if (queue_idx == 2) {
460 if (handle_rx_kick(core, vnet_state) == -1){
461 PrintError("Could not handle Virtio RX buffer refills Kick\n");
464 vnet_state->rx_exit ++;
466 PrintError("VNET Bridge: Kick on invalid queue (%d)\n", queue_idx);
472 case VIRTIO_STATUS_PORT:
473 vnet_state->virtio_cfg.status = *(uint8_t *)src;
475 if (vnet_state->virtio_cfg.status == 0) {
476 PrintDebug("VNET Bridge: Resetting device\n");
477 virtio_reset(vnet_state);
482 case VIRTIO_ISR_PORT:
483 vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
494 static int vnet_virtio_io_read(struct guest_info * core,
495 uint16_t port, void * dst,
496 uint_t length, void * private_data) {
498 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
499 int port_idx = port % vnet_state->io_range_size;
502 case HOST_FEATURES_PORT:
504 PrintError("Illegal read length for host features\n");
508 *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
511 case VRING_PG_NUM_PORT:
513 PrintError("Illegal read length for page frame number\n");
517 *(uint32_t *)dst = vnet_state->cur_queue->pfn;
520 case VRING_SIZE_PORT:
522 PrintError("Illegal read length for vring size\n");
526 *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
530 case VIRTIO_STATUS_PORT:
532 PrintError("Illegal read length for status\n");
536 *(uint8_t *)dst = vnet_state->virtio_cfg.status;
539 case VIRTIO_ISR_PORT:
540 *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
541 vnet_state->virtio_cfg.pci_isr = 0;
542 v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
546 if ( (port_idx >= sizeof(struct virtio_config)) &&
547 (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
548 int cfg_offset = port_idx - sizeof(struct virtio_config);
549 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
551 memcpy(dst, cfg_ptr + cfg_offset, length);
554 PrintError("Read of Unhandled Virtio Read\n");
565 static int virtio_free(struct virtio_vnet_state * vnet_state) {
567 // unregister from PCI
574 static struct v3_device_ops dev_ops = {
575 .free = (int (*)(void *))virtio_free,
578 static int dev_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
579 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
580 struct virtio_vnet_state * vnet_state = NULL;
581 struct pci_device * pci_dev = NULL;
582 char * name = v3_cfg_val(cfg, "name");
584 PrintDebug("VNET Bridge: Initializing VNET Bridge Control device: %s\n", name);
586 if (pci_bus == NULL) {
587 PrintError("VNET Bridge device require a PCI Bus");
591 vnet_state = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
592 memset(vnet_state, 0, sizeof(struct virtio_vnet_state));
596 struct vm_device * dev = v3_add_device(vm, name, &dev_ops, vnet_state);
599 PrintError("Could not attach device %s\n", name);
605 // PCI initialization
607 struct v3_pci_bar bars[6];
608 int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
609 int tmp_ports = num_ports;
612 // This gets the number of ports, rounded up to a power of 2
613 vnet_state->io_range_size = 1; // must be a power of 2
615 while (tmp_ports > 0) {
617 vnet_state->io_range_size <<= 1;
620 // this is to account for any low order bits being set in num_ports
621 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
622 if ((num_ports & ((vnet_state->io_range_size >> 1) - 1)) == 0) {
623 vnet_state->io_range_size >>= 1;
626 for (i = 0; i < 6; i++) {
627 bars[i].type = PCI_BAR_NONE;
630 bars[0].type = PCI_BAR_IO;
631 bars[0].default_base_port = -1;
632 bars[0].num_ports = vnet_state->io_range_size;
633 bars[0].io_read = vnet_virtio_io_read;
634 bars[0].io_write = vnet_virtio_io_write;
635 bars[0].private_data = vnet_state;
637 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
638 0, 5 /*PCI_AUTO_DEV_NUM*/, 0,
639 "LNX_VIRTIO_VNET", bars,
640 NULL, NULL, NULL, vnet_state);
643 PrintError("Could not register PCI Device\n");
644 v3_remove_device(dev);
648 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
649 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
650 pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
651 pci_dev->config_header.class = PCI_CLASS_MEMORY;
652 pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
653 pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
654 pci_dev->config_header.intr_pin = 1;
655 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
658 vnet_state->pci_dev = pci_dev;
659 vnet_state->pci_bus = pci_bus;
662 virtio_reset(vnet_state);
664 struct v3_vnet_bridge_ops brg_ops;
665 brg_ops.input = vnet_pkt_input_cb;
666 brg_ops.poll = vnet_virtio_poll;
668 V3_Print("Registering Virtio device as vnet bridge\n");
670 v3_vnet_add_bridge(vm, &brg_ops, CTL_VM_BRIDGE, (void *)vnet_state);
676 device_register("LNX_VIRTIO_VNET", dev_init)